Communication and congestion aware run-time task mapping on heterogeneous MPSoCs

The proper mapping of applications and their corresponding tasks has a significant effect on both performance and energy consumption in NoC-based MPSoC platforms. It has been shown in the literature that mapping connected tasks onto neighboring processors can control energy consumption and improve performance. This approach reduces the distance between connected tasks, which in-turn decreases the network traffic. This results in a reduction in overall energy consumption. Depending on the application, communication traffic may still cause network congestion which can still significantly affect the overall performance. Most existing methods do not simultaneously consider both the task communications and network congestion. In this paper, these two values together determine the proper mapping of applications on heterogeneous MPSoCs. The Best Neighbor (BN) algorithm is used to determine the initial mapping. Then during run-time, the mapping is adjusted dynamically using the acquired information from network traffic. The objective in the proposed scheme is to map tasks that have to communicate with each other in adjacent nodes, unless the current link traffic is preventing it. Experimental results indicate that the proposed algorithm achieves 23% communication energy saving compared to the BN mapping scheme, and 15% in comparison to the existing communication energy-aware (CE) method for an H.264/AVC decoder mapped on a heterogeneous MPSoC.